Skip to content

Navigation Menu

Sign in
Appearance settings

Search code, repositories, users, issues, pull requests...

Provide feedback

We read every piece of feedback, and take your input very seriously.

Saved searches

Use saved searches to filter your results more quickly

Sign up
Appearance settings

LLM Latency Spikes (3-5 min) during Error Handling in Google ADK #973

Unanswered
fedorovychh asked this question in Q&A
Discussion options

When a tool wrote with Google ADK returns a "failed operation" response, the LLM experiences an abnormal delay (3–5 minutes) before attempting to fix parameters or respond. This "infinite thinking" loop occurs specifically after an error is injected into the conversation context.

What are the most effective strategies to prevent the LLM from getting stuck in a "reasoning loop" after a tool failure—is it better to strip the error response to a bare code, or is there a specific ADK configuration to bypass this latency?

You must be logged in to vote

Replies: 1 comment 1 reply

Comment options

From my point of view, the long pause after a tool failure usually means the model is being given too much ambiguous error context and is trying to reason its way out of an ill-posed repair loop. Bare error codes alone are often too little, but raw stack-like failure prose is often too much.

The most effective pattern is usually a short structured error contract with retryability, suggested parameter corrections, and a hard cap on repair attempts so the agent cannot think in circles for minutes.

You must be logged in to vote
1 reply
Comment options

From my point of view, the long pause after a tool failure usually means the model is being given too much ambiguous error context and is trying to reason its way out of an ill-posed repair loop. Bare error codes alone are often too little, but raw stack-like failure prose is often too much.

The most effective pattern is usually a short structured error contract with retryability, suggested parameter corrections, and a hard cap on repair attempts so the agent cannot think in circles for minutes.

Okay, I'm going to test few cases and debug to find out how much context LLM gets after tool returns error response. It definitely can be the solution to properly handle errors.

I will let you know after getting results. Thank you for the response!

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Category
Q&A
Labels
None yet

AltStyle によって変換されたページ (->オリジナル) /